Harnessing your business' data

If you have an E-commerce business, chances are that you have precious dataset that big marketing platforms don't.
This data can be used to launch a successful, well-targeted marketing campaign.

If you have 10000 customers; you can either, craft an advertisement for each customer, or, you can group customers that are alike and the craft an advertisement for each group; so instead of creating 10000 ads, you'll craft four or eight.

One of the models that are used for this purpose is the RFM model, it stands for:

• Recency: Grouping by when was the customer last order.
  Examples:
  Note: To simplify the examples, H means high value and L means low value.
  

  Customer ID	Recency	Recency grouping
  68426	12	H
  14628	45	H
  56144	167	L
  94247	378	L

• Frequency: Grouping by how many orders a customer has purchased.
  

  Customer ID	Frequency	Frequency grouping
  68426	1	L
  14628	3	H
  56144	2	H
  94247	1	L

• Monetary value: Grouping by how much money a customer did spend.
  

  Customer ID	Monetary value	Monetary value grouping
  68426	20	L
  14628	90	H
  56144	40	L
  94247	120	H

We will have a case study below further discussing what we can do with the grouping/clustering.

As long as our methods rely on statistic and common sense, we can chose multiple rankings and not just high and low.
But, for a start we can use the mean \(\mu\) as a reference point.
Example: a customer with a value that is above mean if the variable can be considered high(H).

We will mimic the e-commerce dataset with this public one Online Retail II by Daqing Chen.
We will be cleaning and manipulating the data to prepare it for the RFM model using logarithmic \(\log\) and normalization \(\frac{x-\mu}{\sigma} transformations\)

The final data that we will be used to make marketing decision can be found Link.
Note: As it is stated above, all the data got scaled using logarithmic and normalize transformations.
Note: There are other methods that can be used like the K-means, but, we will be using the mean \(\mu\) as cluster point for easy explanation.

• Variables in the data:
  
  • Customer_ID: Customer ID
    
  • Time_since_last_invoice: Same as recency (When was the last time a customer ordered?)
    
  • Invoice_Count: Same as frequency (How many times a customer did ordered?)
    
  • Total_Spent: Same as monetary value (How much money a customer spent in total?)
    
  • Cluster1: Customers cluster around one variable (Total_spent)
    
  • Cluster2: Customers clutter around two variables (Total_spent and Time_since_last_invoice)
    
  • Cluster3: Customers clutter around three variable (Total_spent, Time_since_last_invoice, and Invoice_Count)
    

Yes

Yes, instead of H and L, we can have 1, 2, and 3 up to 10 if you desire.
Using the K-means algorithm, there is a method called the elbow method, it determines what is the optimal number of rank to use in a giving data set.